1 /* 2 * QEMU sPAPR PCI host originated from Uninorth PCI host 3 * 4 * Copyright (c) 2011 Alexey Kardashevskiy, IBM Corporation. 5 * Copyright (C) 2011 David Gibson, IBM Corporation. 6 * 7 * Permission is hereby granted, free of charge, to any person obtaining a copy 8 * of this software and associated documentation files (the "Software"), to deal 9 * in the Software without restriction, including without limitation the rights 10 * to use, copy, modify, merge, publish, distribute, sublicense, and/or sell 11 * copies of the Software, and to permit persons to whom the Software is 12 * furnished to do so, subject to the following conditions: 13 * 14 * The above copyright notice and this permission notice shall be included in 15 * all copies or substantial portions of the Software. 16 * 17 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR 18 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY, 19 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL 20 * THE AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER 21 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, 22 * OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN 23 * THE SOFTWARE. 24 */ 25 #include "qemu/osdep.h" 26 #include "qapi/error.h" 27 #include "qemu-common.h" 28 #include "cpu.h" 29 #include "hw/hw.h" 30 #include "hw/sysbus.h" 31 #include "hw/pci/pci.h" 32 #include "hw/pci/msi.h" 33 #include "hw/pci/msix.h" 34 #include "hw/pci/pci_host.h" 35 #include "hw/ppc/spapr.h" 36 #include "hw/pci-host/spapr.h" 37 #include "exec/address-spaces.h" 38 #include "exec/ram_addr.h" 39 #include <libfdt.h> 40 #include "trace.h" 41 #include "qemu/error-report.h" 42 #include "qapi/qmp/qerror.h" 43 #include "hw/ppc/fdt.h" 44 #include "hw/pci/pci_bridge.h" 45 #include "hw/pci/pci_bus.h" 46 #include "hw/pci/pci_ids.h" 47 #include "hw/ppc/spapr_drc.h" 48 #include "sysemu/device_tree.h" 49 #include "sysemu/kvm.h" 50 #include "sysemu/hostmem.h" 51 #include "sysemu/numa.h" 52 53 /* Copied from the kernel arch/powerpc/platforms/pseries/msi.c */ 54 #define RTAS_QUERY_FN 0 55 #define RTAS_CHANGE_FN 1 56 #define RTAS_RESET_FN 2 57 #define RTAS_CHANGE_MSI_FN 3 58 #define RTAS_CHANGE_MSIX_FN 4 59 60 /* Interrupt types to return on RTAS_CHANGE_* */ 61 #define RTAS_TYPE_MSI 1 62 #define RTAS_TYPE_MSIX 2 63 64 SpaprPhbState *spapr_pci_find_phb(SpaprMachineState *spapr, uint64_t buid) 65 { 66 SpaprPhbState *sphb; 67 68 QLIST_FOREACH(sphb, &spapr->phbs, list) { 69 if (sphb->buid != buid) { 70 continue; 71 } 72 return sphb; 73 } 74 75 return NULL; 76 } 77 78 PCIDevice *spapr_pci_find_dev(SpaprMachineState *spapr, uint64_t buid, 79 uint32_t config_addr) 80 { 81 SpaprPhbState *sphb = spapr_pci_find_phb(spapr, buid); 82 PCIHostState *phb = PCI_HOST_BRIDGE(sphb); 83 int bus_num = (config_addr >> 16) & 0xFF; 84 int devfn = (config_addr >> 8) & 0xFF; 85 86 if (!phb) { 87 return NULL; 88 } 89 90 return pci_find_device(phb->bus, bus_num, devfn); 91 } 92 93 static uint32_t rtas_pci_cfgaddr(uint32_t arg) 94 { 95 /* This handles the encoding of extended config space addresses */ 96 return ((arg >> 20) & 0xf00) | (arg & 0xff); 97 } 98 99 static void finish_read_pci_config(SpaprMachineState *spapr, uint64_t buid, 100 uint32_t addr, uint32_t size, 101 target_ulong rets) 102 { 103 PCIDevice *pci_dev; 104 uint32_t val; 105 106 if ((size != 1) && (size != 2) && (size != 4)) { 107 /* access must be 1, 2 or 4 bytes */ 108 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 109 return; 110 } 111 112 pci_dev = spapr_pci_find_dev(spapr, buid, addr); 113 addr = rtas_pci_cfgaddr(addr); 114 115 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) { 116 /* Access must be to a valid device, within bounds and 117 * naturally aligned */ 118 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 119 return; 120 } 121 122 val = pci_host_config_read_common(pci_dev, addr, 123 pci_config_size(pci_dev), size); 124 125 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 126 rtas_st(rets, 1, val); 127 } 128 129 static void rtas_ibm_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 130 uint32_t token, uint32_t nargs, 131 target_ulong args, 132 uint32_t nret, target_ulong rets) 133 { 134 uint64_t buid; 135 uint32_t size, addr; 136 137 if ((nargs != 4) || (nret != 2)) { 138 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 139 return; 140 } 141 142 buid = rtas_ldq(args, 1); 143 size = rtas_ld(args, 3); 144 addr = rtas_ld(args, 0); 145 146 finish_read_pci_config(spapr, buid, addr, size, rets); 147 } 148 149 static void rtas_read_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 150 uint32_t token, uint32_t nargs, 151 target_ulong args, 152 uint32_t nret, target_ulong rets) 153 { 154 uint32_t size, addr; 155 156 if ((nargs != 2) || (nret != 2)) { 157 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 158 return; 159 } 160 161 size = rtas_ld(args, 1); 162 addr = rtas_ld(args, 0); 163 164 finish_read_pci_config(spapr, 0, addr, size, rets); 165 } 166 167 static void finish_write_pci_config(SpaprMachineState *spapr, uint64_t buid, 168 uint32_t addr, uint32_t size, 169 uint32_t val, target_ulong rets) 170 { 171 PCIDevice *pci_dev; 172 173 if ((size != 1) && (size != 2) && (size != 4)) { 174 /* access must be 1, 2 or 4 bytes */ 175 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 176 return; 177 } 178 179 pci_dev = spapr_pci_find_dev(spapr, buid, addr); 180 addr = rtas_pci_cfgaddr(addr); 181 182 if (!pci_dev || (addr % size) || (addr >= pci_config_size(pci_dev))) { 183 /* Access must be to a valid device, within bounds and 184 * naturally aligned */ 185 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 186 return; 187 } 188 189 pci_host_config_write_common(pci_dev, addr, pci_config_size(pci_dev), 190 val, size); 191 192 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 193 } 194 195 static void rtas_ibm_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 196 uint32_t token, uint32_t nargs, 197 target_ulong args, 198 uint32_t nret, target_ulong rets) 199 { 200 uint64_t buid; 201 uint32_t val, size, addr; 202 203 if ((nargs != 5) || (nret != 1)) { 204 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 205 return; 206 } 207 208 buid = rtas_ldq(args, 1); 209 val = rtas_ld(args, 4); 210 size = rtas_ld(args, 3); 211 addr = rtas_ld(args, 0); 212 213 finish_write_pci_config(spapr, buid, addr, size, val, rets); 214 } 215 216 static void rtas_write_pci_config(PowerPCCPU *cpu, SpaprMachineState *spapr, 217 uint32_t token, uint32_t nargs, 218 target_ulong args, 219 uint32_t nret, target_ulong rets) 220 { 221 uint32_t val, size, addr; 222 223 if ((nargs != 3) || (nret != 1)) { 224 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 225 return; 226 } 227 228 229 val = rtas_ld(args, 2); 230 size = rtas_ld(args, 1); 231 addr = rtas_ld(args, 0); 232 233 finish_write_pci_config(spapr, 0, addr, size, val, rets); 234 } 235 236 /* 237 * Set MSI/MSIX message data. 238 * This is required for msi_notify()/msix_notify() which 239 * will write at the addresses via spapr_msi_write(). 240 * 241 * If hwaddr == 0, all entries will have .data == first_irq i.e. 242 * table will be reset. 243 */ 244 static void spapr_msi_setmsg(PCIDevice *pdev, hwaddr addr, bool msix, 245 unsigned first_irq, unsigned req_num) 246 { 247 unsigned i; 248 MSIMessage msg = { .address = addr, .data = first_irq }; 249 250 if (!msix) { 251 msi_set_message(pdev, msg); 252 trace_spapr_pci_msi_setup(pdev->name, 0, msg.address); 253 return; 254 } 255 256 for (i = 0; i < req_num; ++i) { 257 msix_set_message(pdev, i, msg); 258 trace_spapr_pci_msi_setup(pdev->name, i, msg.address); 259 if (addr) { 260 ++msg.data; 261 } 262 } 263 } 264 265 static void rtas_ibm_change_msi(PowerPCCPU *cpu, SpaprMachineState *spapr, 266 uint32_t token, uint32_t nargs, 267 target_ulong args, uint32_t nret, 268 target_ulong rets) 269 { 270 SpaprMachineClass *smc = SPAPR_MACHINE_GET_CLASS(spapr); 271 uint32_t config_addr = rtas_ld(args, 0); 272 uint64_t buid = rtas_ldq(args, 1); 273 unsigned int func = rtas_ld(args, 3); 274 unsigned int req_num = rtas_ld(args, 4); /* 0 == remove all */ 275 unsigned int seq_num = rtas_ld(args, 5); 276 unsigned int ret_intr_type; 277 unsigned int irq, max_irqs = 0; 278 SpaprPhbState *phb = NULL; 279 PCIDevice *pdev = NULL; 280 spapr_pci_msi *msi; 281 int *config_addr_key; 282 Error *err = NULL; 283 int i; 284 285 /* Fins SpaprPhbState */ 286 phb = spapr_pci_find_phb(spapr, buid); 287 if (phb) { 288 pdev = spapr_pci_find_dev(spapr, buid, config_addr); 289 } 290 if (!phb || !pdev) { 291 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 292 return; 293 } 294 295 switch (func) { 296 case RTAS_CHANGE_FN: 297 if (msi_present(pdev)) { 298 ret_intr_type = RTAS_TYPE_MSI; 299 } else if (msix_present(pdev)) { 300 ret_intr_type = RTAS_TYPE_MSIX; 301 } else { 302 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 303 return; 304 } 305 break; 306 case RTAS_CHANGE_MSI_FN: 307 if (msi_present(pdev)) { 308 ret_intr_type = RTAS_TYPE_MSI; 309 } else { 310 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 311 return; 312 } 313 break; 314 case RTAS_CHANGE_MSIX_FN: 315 if (msix_present(pdev)) { 316 ret_intr_type = RTAS_TYPE_MSIX; 317 } else { 318 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 319 return; 320 } 321 break; 322 default: 323 error_report("rtas_ibm_change_msi(%u) is not implemented", func); 324 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 325 return; 326 } 327 328 msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr); 329 330 /* Releasing MSIs */ 331 if (!req_num) { 332 if (!msi) { 333 trace_spapr_pci_msi("Releasing wrong config", config_addr); 334 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 335 return; 336 } 337 338 if (!smc->legacy_irq_allocation) { 339 spapr_irq_msi_free(spapr, msi->first_irq, msi->num); 340 } 341 spapr_irq_free(spapr, msi->first_irq, msi->num); 342 if (msi_present(pdev)) { 343 spapr_msi_setmsg(pdev, 0, false, 0, 0); 344 } 345 if (msix_present(pdev)) { 346 spapr_msi_setmsg(pdev, 0, true, 0, 0); 347 } 348 g_hash_table_remove(phb->msi, &config_addr); 349 350 trace_spapr_pci_msi("Released MSIs", config_addr); 351 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 352 rtas_st(rets, 1, 0); 353 return; 354 } 355 356 /* Enabling MSI */ 357 358 /* Check if the device supports as many IRQs as requested */ 359 if (ret_intr_type == RTAS_TYPE_MSI) { 360 max_irqs = msi_nr_vectors_allocated(pdev); 361 } else if (ret_intr_type == RTAS_TYPE_MSIX) { 362 max_irqs = pdev->msix_entries_nr; 363 } 364 if (!max_irqs) { 365 error_report("Requested interrupt type %d is not enabled for device %x", 366 ret_intr_type, config_addr); 367 rtas_st(rets, 0, -1); /* Hardware error */ 368 return; 369 } 370 /* Correct the number if the guest asked for too many */ 371 if (req_num > max_irqs) { 372 trace_spapr_pci_msi_retry(config_addr, req_num, max_irqs); 373 req_num = max_irqs; 374 irq = 0; /* to avoid misleading trace */ 375 goto out; 376 } 377 378 /* Allocate MSIs */ 379 if (smc->legacy_irq_allocation) { 380 irq = spapr_irq_find(spapr, req_num, ret_intr_type == RTAS_TYPE_MSI, 381 &err); 382 } else { 383 irq = spapr_irq_msi_alloc(spapr, req_num, 384 ret_intr_type == RTAS_TYPE_MSI, &err); 385 } 386 if (err) { 387 error_reportf_err(err, "Can't allocate MSIs for device %x: ", 388 config_addr); 389 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 390 return; 391 } 392 393 for (i = 0; i < req_num; i++) { 394 spapr_irq_claim(spapr, irq + i, false, &err); 395 if (err) { 396 if (i) { 397 spapr_irq_free(spapr, irq, i); 398 } 399 if (!smc->legacy_irq_allocation) { 400 spapr_irq_msi_free(spapr, irq, req_num); 401 } 402 error_reportf_err(err, "Can't allocate MSIs for device %x: ", 403 config_addr); 404 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 405 return; 406 } 407 } 408 409 /* Release previous MSIs */ 410 if (msi) { 411 if (!smc->legacy_irq_allocation) { 412 spapr_irq_msi_free(spapr, msi->first_irq, msi->num); 413 } 414 spapr_irq_free(spapr, msi->first_irq, msi->num); 415 g_hash_table_remove(phb->msi, &config_addr); 416 } 417 418 /* Setup MSI/MSIX vectors in the device (via cfgspace or MSIX BAR) */ 419 spapr_msi_setmsg(pdev, SPAPR_PCI_MSI_WINDOW, ret_intr_type == RTAS_TYPE_MSIX, 420 irq, req_num); 421 422 /* Add MSI device to cache */ 423 msi = g_new(spapr_pci_msi, 1); 424 msi->first_irq = irq; 425 msi->num = req_num; 426 config_addr_key = g_new(int, 1); 427 *config_addr_key = config_addr; 428 g_hash_table_insert(phb->msi, config_addr_key, msi); 429 430 out: 431 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 432 rtas_st(rets, 1, req_num); 433 rtas_st(rets, 2, ++seq_num); 434 if (nret > 3) { 435 rtas_st(rets, 3, ret_intr_type); 436 } 437 438 trace_spapr_pci_rtas_ibm_change_msi(config_addr, func, req_num, irq); 439 } 440 441 static void rtas_ibm_query_interrupt_source_number(PowerPCCPU *cpu, 442 SpaprMachineState *spapr, 443 uint32_t token, 444 uint32_t nargs, 445 target_ulong args, 446 uint32_t nret, 447 target_ulong rets) 448 { 449 uint32_t config_addr = rtas_ld(args, 0); 450 uint64_t buid = rtas_ldq(args, 1); 451 unsigned int intr_src_num = -1, ioa_intr_num = rtas_ld(args, 3); 452 SpaprPhbState *phb = NULL; 453 PCIDevice *pdev = NULL; 454 spapr_pci_msi *msi; 455 456 /* Find SpaprPhbState */ 457 phb = spapr_pci_find_phb(spapr, buid); 458 if (phb) { 459 pdev = spapr_pci_find_dev(spapr, buid, config_addr); 460 } 461 if (!phb || !pdev) { 462 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 463 return; 464 } 465 466 /* Find device descriptor and start IRQ */ 467 msi = (spapr_pci_msi *) g_hash_table_lookup(phb->msi, &config_addr); 468 if (!msi || !msi->first_irq || !msi->num || (ioa_intr_num >= msi->num)) { 469 trace_spapr_pci_msi("Failed to return vector", config_addr); 470 rtas_st(rets, 0, RTAS_OUT_HW_ERROR); 471 return; 472 } 473 intr_src_num = msi->first_irq + ioa_intr_num; 474 trace_spapr_pci_rtas_ibm_query_interrupt_source_number(ioa_intr_num, 475 intr_src_num); 476 477 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 478 rtas_st(rets, 1, intr_src_num); 479 rtas_st(rets, 2, 1);/* 0 == level; 1 == edge */ 480 } 481 482 static void rtas_ibm_set_eeh_option(PowerPCCPU *cpu, 483 SpaprMachineState *spapr, 484 uint32_t token, uint32_t nargs, 485 target_ulong args, uint32_t nret, 486 target_ulong rets) 487 { 488 SpaprPhbState *sphb; 489 uint32_t addr, option; 490 uint64_t buid; 491 int ret; 492 493 if ((nargs != 4) || (nret != 1)) { 494 goto param_error_exit; 495 } 496 497 buid = rtas_ldq(args, 1); 498 addr = rtas_ld(args, 0); 499 option = rtas_ld(args, 3); 500 501 sphb = spapr_pci_find_phb(spapr, buid); 502 if (!sphb) { 503 goto param_error_exit; 504 } 505 506 if (!spapr_phb_eeh_available(sphb)) { 507 goto param_error_exit; 508 } 509 510 ret = spapr_phb_vfio_eeh_set_option(sphb, addr, option); 511 rtas_st(rets, 0, ret); 512 return; 513 514 param_error_exit: 515 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 516 } 517 518 static void rtas_ibm_get_config_addr_info2(PowerPCCPU *cpu, 519 SpaprMachineState *spapr, 520 uint32_t token, uint32_t nargs, 521 target_ulong args, uint32_t nret, 522 target_ulong rets) 523 { 524 SpaprPhbState *sphb; 525 PCIDevice *pdev; 526 uint32_t addr, option; 527 uint64_t buid; 528 529 if ((nargs != 4) || (nret != 2)) { 530 goto param_error_exit; 531 } 532 533 buid = rtas_ldq(args, 1); 534 sphb = spapr_pci_find_phb(spapr, buid); 535 if (!sphb) { 536 goto param_error_exit; 537 } 538 539 if (!spapr_phb_eeh_available(sphb)) { 540 goto param_error_exit; 541 } 542 543 /* 544 * We always have PE address of form "00BB0001". "BB" 545 * represents the bus number of PE's primary bus. 546 */ 547 option = rtas_ld(args, 3); 548 switch (option) { 549 case RTAS_GET_PE_ADDR: 550 addr = rtas_ld(args, 0); 551 pdev = spapr_pci_find_dev(spapr, buid, addr); 552 if (!pdev) { 553 goto param_error_exit; 554 } 555 556 rtas_st(rets, 1, (pci_bus_num(pci_get_bus(pdev)) << 16) + 1); 557 break; 558 case RTAS_GET_PE_MODE: 559 rtas_st(rets, 1, RTAS_PE_MODE_SHARED); 560 break; 561 default: 562 goto param_error_exit; 563 } 564 565 rtas_st(rets, 0, RTAS_OUT_SUCCESS); 566 return; 567 568 param_error_exit: 569 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 570 } 571 572 static void rtas_ibm_read_slot_reset_state2(PowerPCCPU *cpu, 573 SpaprMachineState *spapr, 574 uint32_t token, uint32_t nargs, 575 target_ulong args, uint32_t nret, 576 target_ulong rets) 577 { 578 SpaprPhbState *sphb; 579 uint64_t buid; 580 int state, ret; 581 582 if ((nargs != 3) || (nret != 4 && nret != 5)) { 583 goto param_error_exit; 584 } 585 586 buid = rtas_ldq(args, 1); 587 sphb = spapr_pci_find_phb(spapr, buid); 588 if (!sphb) { 589 goto param_error_exit; 590 } 591 592 if (!spapr_phb_eeh_available(sphb)) { 593 goto param_error_exit; 594 } 595 596 ret = spapr_phb_vfio_eeh_get_state(sphb, &state); 597 rtas_st(rets, 0, ret); 598 if (ret != RTAS_OUT_SUCCESS) { 599 return; 600 } 601 602 rtas_st(rets, 1, state); 603 rtas_st(rets, 2, RTAS_EEH_SUPPORT); 604 rtas_st(rets, 3, RTAS_EEH_PE_UNAVAIL_INFO); 605 if (nret >= 5) { 606 rtas_st(rets, 4, RTAS_EEH_PE_RECOVER_INFO); 607 } 608 return; 609 610 param_error_exit: 611 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 612 } 613 614 static void rtas_ibm_set_slot_reset(PowerPCCPU *cpu, 615 SpaprMachineState *spapr, 616 uint32_t token, uint32_t nargs, 617 target_ulong args, uint32_t nret, 618 target_ulong rets) 619 { 620 SpaprPhbState *sphb; 621 uint32_t option; 622 uint64_t buid; 623 int ret; 624 625 if ((nargs != 4) || (nret != 1)) { 626 goto param_error_exit; 627 } 628 629 buid = rtas_ldq(args, 1); 630 option = rtas_ld(args, 3); 631 sphb = spapr_pci_find_phb(spapr, buid); 632 if (!sphb) { 633 goto param_error_exit; 634 } 635 636 if (!spapr_phb_eeh_available(sphb)) { 637 goto param_error_exit; 638 } 639 640 ret = spapr_phb_vfio_eeh_reset(sphb, option); 641 rtas_st(rets, 0, ret); 642 return; 643 644 param_error_exit: 645 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 646 } 647 648 static void rtas_ibm_configure_pe(PowerPCCPU *cpu, 649 SpaprMachineState *spapr, 650 uint32_t token, uint32_t nargs, 651 target_ulong args, uint32_t nret, 652 target_ulong rets) 653 { 654 SpaprPhbState *sphb; 655 uint64_t buid; 656 int ret; 657 658 if ((nargs != 3) || (nret != 1)) { 659 goto param_error_exit; 660 } 661 662 buid = rtas_ldq(args, 1); 663 sphb = spapr_pci_find_phb(spapr, buid); 664 if (!sphb) { 665 goto param_error_exit; 666 } 667 668 if (!spapr_phb_eeh_available(sphb)) { 669 goto param_error_exit; 670 } 671 672 ret = spapr_phb_vfio_eeh_configure(sphb); 673 rtas_st(rets, 0, ret); 674 return; 675 676 param_error_exit: 677 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 678 } 679 680 /* To support it later */ 681 static void rtas_ibm_slot_error_detail(PowerPCCPU *cpu, 682 SpaprMachineState *spapr, 683 uint32_t token, uint32_t nargs, 684 target_ulong args, uint32_t nret, 685 target_ulong rets) 686 { 687 SpaprPhbState *sphb; 688 int option; 689 uint64_t buid; 690 691 if ((nargs != 8) || (nret != 1)) { 692 goto param_error_exit; 693 } 694 695 buid = rtas_ldq(args, 1); 696 sphb = spapr_pci_find_phb(spapr, buid); 697 if (!sphb) { 698 goto param_error_exit; 699 } 700 701 if (!spapr_phb_eeh_available(sphb)) { 702 goto param_error_exit; 703 } 704 705 option = rtas_ld(args, 7); 706 switch (option) { 707 case RTAS_SLOT_TEMP_ERR_LOG: 708 case RTAS_SLOT_PERM_ERR_LOG: 709 break; 710 default: 711 goto param_error_exit; 712 } 713 714 /* We don't have error log yet */ 715 rtas_st(rets, 0, RTAS_OUT_NO_ERRORS_FOUND); 716 return; 717 718 param_error_exit: 719 rtas_st(rets, 0, RTAS_OUT_PARAM_ERROR); 720 } 721 722 static void pci_spapr_set_irq(void *opaque, int irq_num, int level) 723 { 724 /* 725 * Here we use the number returned by pci_swizzle_map_irq_fn to find a 726 * corresponding qemu_irq. 727 */ 728 SpaprPhbState *phb = opaque; 729 730 trace_spapr_pci_lsi_set(phb->dtbusname, irq_num, phb->lsi_table[irq_num].irq); 731 qemu_set_irq(spapr_phb_lsi_qirq(phb, irq_num), level); 732 } 733 734 static PCIINTxRoute spapr_route_intx_pin_to_irq(void *opaque, int pin) 735 { 736 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(opaque); 737 PCIINTxRoute route; 738 739 route.mode = PCI_INTX_ENABLED; 740 route.irq = sphb->lsi_table[pin].irq; 741 742 return route; 743 } 744 745 /* 746 * MSI/MSIX memory region implementation. 747 * The handler handles both MSI and MSIX. 748 * The vector number is encoded in least bits in data. 749 */ 750 static void spapr_msi_write(void *opaque, hwaddr addr, 751 uint64_t data, unsigned size) 752 { 753 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 754 uint32_t irq = data; 755 756 trace_spapr_pci_msi_write(addr, data, irq); 757 758 qemu_irq_pulse(spapr_qirq(spapr, irq)); 759 } 760 761 static const MemoryRegionOps spapr_msi_ops = { 762 /* There is no .read as the read result is undefined by PCI spec */ 763 .read = NULL, 764 .write = spapr_msi_write, 765 .endianness = DEVICE_LITTLE_ENDIAN 766 }; 767 768 /* 769 * PHB PCI device 770 */ 771 static AddressSpace *spapr_pci_dma_iommu(PCIBus *bus, void *opaque, int devfn) 772 { 773 SpaprPhbState *phb = opaque; 774 775 return &phb->iommu_as; 776 } 777 778 static char *spapr_phb_vfio_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev) 779 { 780 char *path = NULL, *buf = NULL, *host = NULL; 781 782 /* Get the PCI VFIO host id */ 783 host = object_property_get_str(OBJECT(pdev), "host", NULL); 784 if (!host) { 785 goto err_out; 786 } 787 788 /* Construct the path of the file that will give us the DT location */ 789 path = g_strdup_printf("/sys/bus/pci/devices/%s/devspec", host); 790 g_free(host); 791 if (!g_file_get_contents(path, &buf, NULL, NULL)) { 792 goto err_out; 793 } 794 g_free(path); 795 796 /* Construct and read from host device tree the loc-code */ 797 path = g_strdup_printf("/proc/device-tree%s/ibm,loc-code", buf); 798 g_free(buf); 799 if (!g_file_get_contents(path, &buf, NULL, NULL)) { 800 goto err_out; 801 } 802 return buf; 803 804 err_out: 805 g_free(path); 806 return NULL; 807 } 808 809 static char *spapr_phb_get_loc_code(SpaprPhbState *sphb, PCIDevice *pdev) 810 { 811 char *buf; 812 const char *devtype = "qemu"; 813 uint32_t busnr = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(pdev)))); 814 815 if (object_dynamic_cast(OBJECT(pdev), "vfio-pci")) { 816 buf = spapr_phb_vfio_get_loc_code(sphb, pdev); 817 if (buf) { 818 return buf; 819 } 820 devtype = "vfio"; 821 } 822 /* 823 * For emulated devices and VFIO-failure case, make up 824 * the loc-code. 825 */ 826 buf = g_strdup_printf("%s_%s:%04x:%02x:%02x.%x", 827 devtype, pdev->name, sphb->index, busnr, 828 PCI_SLOT(pdev->devfn), PCI_FUNC(pdev->devfn)); 829 return buf; 830 } 831 832 /* Macros to operate with address in OF binding to PCI */ 833 #define b_x(x, p, l) (((x) & ((1<<(l))-1)) << (p)) 834 #define b_n(x) b_x((x), 31, 1) /* 0 if relocatable */ 835 #define b_p(x) b_x((x), 30, 1) /* 1 if prefetchable */ 836 #define b_t(x) b_x((x), 29, 1) /* 1 if the address is aliased */ 837 #define b_ss(x) b_x((x), 24, 2) /* the space code */ 838 #define b_bbbbbbbb(x) b_x((x), 16, 8) /* bus number */ 839 #define b_ddddd(x) b_x((x), 11, 5) /* device number */ 840 #define b_fff(x) b_x((x), 8, 3) /* function number */ 841 #define b_rrrrrrrr(x) b_x((x), 0, 8) /* register number */ 842 843 /* for 'reg'/'assigned-addresses' OF properties */ 844 #define RESOURCE_CELLS_SIZE 2 845 #define RESOURCE_CELLS_ADDRESS 3 846 847 typedef struct ResourceFields { 848 uint32_t phys_hi; 849 uint32_t phys_mid; 850 uint32_t phys_lo; 851 uint32_t size_hi; 852 uint32_t size_lo; 853 } QEMU_PACKED ResourceFields; 854 855 typedef struct ResourceProps { 856 ResourceFields reg[8]; 857 ResourceFields assigned[7]; 858 uint32_t reg_len; 859 uint32_t assigned_len; 860 } ResourceProps; 861 862 /* fill in the 'reg'/'assigned-resources' OF properties for 863 * a PCI device. 'reg' describes resource requirements for a 864 * device's IO/MEM regions, 'assigned-addresses' describes the 865 * actual resource assignments. 866 * 867 * the properties are arrays of ('phys-addr', 'size') pairs describing 868 * the addressable regions of the PCI device, where 'phys-addr' is a 869 * RESOURCE_CELLS_ADDRESS-tuple of 32-bit integers corresponding to 870 * (phys.hi, phys.mid, phys.lo), and 'size' is a 871 * RESOURCE_CELLS_SIZE-tuple corresponding to (size.hi, size.lo). 872 * 873 * phys.hi = 0xYYXXXXZZ, where: 874 * 0xYY = npt000ss 875 * ||| | 876 * ||| +-- space code 877 * ||| | 878 * ||| + 00 if configuration space 879 * ||| + 01 if IO region, 880 * ||| + 10 if 32-bit MEM region 881 * ||| + 11 if 64-bit MEM region 882 * ||| 883 * ||+------ for non-relocatable IO: 1 if aliased 884 * || for relocatable IO: 1 if below 64KB 885 * || for MEM: 1 if below 1MB 886 * |+------- 1 if region is prefetchable 887 * +-------- 1 if region is non-relocatable 888 * 0xXXXX = bbbbbbbb dddddfff, encoding bus, slot, and function 889 * bits respectively 890 * 0xZZ = rrrrrrrr, the register number of the BAR corresponding 891 * to the region 892 * 893 * phys.mid and phys.lo correspond respectively to the hi/lo portions 894 * of the actual address of the region. 895 * 896 * how the phys-addr/size values are used differ slightly between 897 * 'reg' and 'assigned-addresses' properties. namely, 'reg' has 898 * an additional description for the config space region of the 899 * device, and in the case of QEMU has n=0 and phys.mid=phys.lo=0 900 * to describe the region as relocatable, with an address-mapping 901 * that corresponds directly to the PHB's address space for the 902 * resource. 'assigned-addresses' always has n=1 set with an absolute 903 * address assigned for the resource. in general, 'assigned-addresses' 904 * won't be populated, since addresses for PCI devices are generally 905 * unmapped initially and left to the guest to assign. 906 * 907 * note also that addresses defined in these properties are, at least 908 * for PAPR guests, relative to the PHBs IO/MEM windows, and 909 * correspond directly to the addresses in the BARs. 910 * 911 * in accordance with PCI Bus Binding to Open Firmware, 912 * IEEE Std 1275-1994, section 4.1.1, as implemented by PAPR+ v2.7, 913 * Appendix C. 914 */ 915 static void populate_resource_props(PCIDevice *d, ResourceProps *rp) 916 { 917 int bus_num = pci_bus_num(PCI_BUS(qdev_get_parent_bus(DEVICE(d)))); 918 uint32_t dev_id = (b_bbbbbbbb(bus_num) | 919 b_ddddd(PCI_SLOT(d->devfn)) | 920 b_fff(PCI_FUNC(d->devfn))); 921 ResourceFields *reg, *assigned; 922 int i, reg_idx = 0, assigned_idx = 0; 923 924 /* config space region */ 925 reg = &rp->reg[reg_idx++]; 926 reg->phys_hi = cpu_to_be32(dev_id); 927 reg->phys_mid = 0; 928 reg->phys_lo = 0; 929 reg->size_hi = 0; 930 reg->size_lo = 0; 931 932 for (i = 0; i < PCI_NUM_REGIONS; i++) { 933 if (!d->io_regions[i].size) { 934 continue; 935 } 936 937 reg = &rp->reg[reg_idx++]; 938 939 reg->phys_hi = cpu_to_be32(dev_id | b_rrrrrrrr(pci_bar(d, i))); 940 if (d->io_regions[i].type & PCI_BASE_ADDRESS_SPACE_IO) { 941 reg->phys_hi |= cpu_to_be32(b_ss(1)); 942 } else if (d->io_regions[i].type & PCI_BASE_ADDRESS_MEM_TYPE_64) { 943 reg->phys_hi |= cpu_to_be32(b_ss(3)); 944 } else { 945 reg->phys_hi |= cpu_to_be32(b_ss(2)); 946 } 947 reg->phys_mid = 0; 948 reg->phys_lo = 0; 949 reg->size_hi = cpu_to_be32(d->io_regions[i].size >> 32); 950 reg->size_lo = cpu_to_be32(d->io_regions[i].size); 951 952 if (d->io_regions[i].addr == PCI_BAR_UNMAPPED) { 953 continue; 954 } 955 956 assigned = &rp->assigned[assigned_idx++]; 957 assigned->phys_hi = cpu_to_be32(be32_to_cpu(reg->phys_hi) | b_n(1)); 958 assigned->phys_mid = cpu_to_be32(d->io_regions[i].addr >> 32); 959 assigned->phys_lo = cpu_to_be32(d->io_regions[i].addr); 960 assigned->size_hi = reg->size_hi; 961 assigned->size_lo = reg->size_lo; 962 } 963 964 rp->reg_len = reg_idx * sizeof(ResourceFields); 965 rp->assigned_len = assigned_idx * sizeof(ResourceFields); 966 } 967 968 typedef struct PCIClass PCIClass; 969 typedef struct PCISubClass PCISubClass; 970 typedef struct PCIIFace PCIIFace; 971 972 struct PCIIFace { 973 int iface; 974 const char *name; 975 }; 976 977 struct PCISubClass { 978 int subclass; 979 const char *name; 980 const PCIIFace *iface; 981 }; 982 983 struct PCIClass { 984 const char *name; 985 const PCISubClass *subc; 986 }; 987 988 static const PCISubClass undef_subclass[] = { 989 { PCI_CLASS_NOT_DEFINED_VGA, "display", NULL }, 990 { 0xFF, NULL, NULL }, 991 }; 992 993 static const PCISubClass mass_subclass[] = { 994 { PCI_CLASS_STORAGE_SCSI, "scsi", NULL }, 995 { PCI_CLASS_STORAGE_IDE, "ide", NULL }, 996 { PCI_CLASS_STORAGE_FLOPPY, "fdc", NULL }, 997 { PCI_CLASS_STORAGE_IPI, "ipi", NULL }, 998 { PCI_CLASS_STORAGE_RAID, "raid", NULL }, 999 { PCI_CLASS_STORAGE_ATA, "ata", NULL }, 1000 { PCI_CLASS_STORAGE_SATA, "sata", NULL }, 1001 { PCI_CLASS_STORAGE_SAS, "sas", NULL }, 1002 { 0xFF, NULL, NULL }, 1003 }; 1004 1005 static const PCISubClass net_subclass[] = { 1006 { PCI_CLASS_NETWORK_ETHERNET, "ethernet", NULL }, 1007 { PCI_CLASS_NETWORK_TOKEN_RING, "token-ring", NULL }, 1008 { PCI_CLASS_NETWORK_FDDI, "fddi", NULL }, 1009 { PCI_CLASS_NETWORK_ATM, "atm", NULL }, 1010 { PCI_CLASS_NETWORK_ISDN, "isdn", NULL }, 1011 { PCI_CLASS_NETWORK_WORLDFIP, "worldfip", NULL }, 1012 { PCI_CLASS_NETWORK_PICMG214, "picmg", NULL }, 1013 { 0xFF, NULL, NULL }, 1014 }; 1015 1016 static const PCISubClass displ_subclass[] = { 1017 { PCI_CLASS_DISPLAY_VGA, "vga", NULL }, 1018 { PCI_CLASS_DISPLAY_XGA, "xga", NULL }, 1019 { PCI_CLASS_DISPLAY_3D, "3d-controller", NULL }, 1020 { 0xFF, NULL, NULL }, 1021 }; 1022 1023 static const PCISubClass media_subclass[] = { 1024 { PCI_CLASS_MULTIMEDIA_VIDEO, "video", NULL }, 1025 { PCI_CLASS_MULTIMEDIA_AUDIO, "sound", NULL }, 1026 { PCI_CLASS_MULTIMEDIA_PHONE, "telephony", NULL }, 1027 { 0xFF, NULL, NULL }, 1028 }; 1029 1030 static const PCISubClass mem_subclass[] = { 1031 { PCI_CLASS_MEMORY_RAM, "memory", NULL }, 1032 { PCI_CLASS_MEMORY_FLASH, "flash", NULL }, 1033 { 0xFF, NULL, NULL }, 1034 }; 1035 1036 static const PCISubClass bridg_subclass[] = { 1037 { PCI_CLASS_BRIDGE_HOST, "host", NULL }, 1038 { PCI_CLASS_BRIDGE_ISA, "isa", NULL }, 1039 { PCI_CLASS_BRIDGE_EISA, "eisa", NULL }, 1040 { PCI_CLASS_BRIDGE_MC, "mca", NULL }, 1041 { PCI_CLASS_BRIDGE_PCI, "pci", NULL }, 1042 { PCI_CLASS_BRIDGE_PCMCIA, "pcmcia", NULL }, 1043 { PCI_CLASS_BRIDGE_NUBUS, "nubus", NULL }, 1044 { PCI_CLASS_BRIDGE_CARDBUS, "cardbus", NULL }, 1045 { PCI_CLASS_BRIDGE_RACEWAY, "raceway", NULL }, 1046 { PCI_CLASS_BRIDGE_PCI_SEMITP, "semi-transparent-pci", NULL }, 1047 { PCI_CLASS_BRIDGE_IB_PCI, "infiniband", NULL }, 1048 { 0xFF, NULL, NULL }, 1049 }; 1050 1051 static const PCISubClass comm_subclass[] = { 1052 { PCI_CLASS_COMMUNICATION_SERIAL, "serial", NULL }, 1053 { PCI_CLASS_COMMUNICATION_PARALLEL, "parallel", NULL }, 1054 { PCI_CLASS_COMMUNICATION_MULTISERIAL, "multiport-serial", NULL }, 1055 { PCI_CLASS_COMMUNICATION_MODEM, "modem", NULL }, 1056 { PCI_CLASS_COMMUNICATION_GPIB, "gpib", NULL }, 1057 { PCI_CLASS_COMMUNICATION_SC, "smart-card", NULL }, 1058 { 0xFF, NULL, NULL, }, 1059 }; 1060 1061 static const PCIIFace pic_iface[] = { 1062 { PCI_CLASS_SYSTEM_PIC_IOAPIC, "io-apic" }, 1063 { PCI_CLASS_SYSTEM_PIC_IOXAPIC, "io-xapic" }, 1064 { 0xFF, NULL }, 1065 }; 1066 1067 static const PCISubClass sys_subclass[] = { 1068 { PCI_CLASS_SYSTEM_PIC, "interrupt-controller", pic_iface }, 1069 { PCI_CLASS_SYSTEM_DMA, "dma-controller", NULL }, 1070 { PCI_CLASS_SYSTEM_TIMER, "timer", NULL }, 1071 { PCI_CLASS_SYSTEM_RTC, "rtc", NULL }, 1072 { PCI_CLASS_SYSTEM_PCI_HOTPLUG, "hot-plug-controller", NULL }, 1073 { PCI_CLASS_SYSTEM_SDHCI, "sd-host-controller", NULL }, 1074 { 0xFF, NULL, NULL }, 1075 }; 1076 1077 static const PCISubClass inp_subclass[] = { 1078 { PCI_CLASS_INPUT_KEYBOARD, "keyboard", NULL }, 1079 { PCI_CLASS_INPUT_PEN, "pen", NULL }, 1080 { PCI_CLASS_INPUT_MOUSE, "mouse", NULL }, 1081 { PCI_CLASS_INPUT_SCANNER, "scanner", NULL }, 1082 { PCI_CLASS_INPUT_GAMEPORT, "gameport", NULL }, 1083 { 0xFF, NULL, NULL }, 1084 }; 1085 1086 static const PCISubClass dock_subclass[] = { 1087 { PCI_CLASS_DOCKING_GENERIC, "dock", NULL }, 1088 { 0xFF, NULL, NULL }, 1089 }; 1090 1091 static const PCISubClass cpu_subclass[] = { 1092 { PCI_CLASS_PROCESSOR_PENTIUM, "pentium", NULL }, 1093 { PCI_CLASS_PROCESSOR_POWERPC, "powerpc", NULL }, 1094 { PCI_CLASS_PROCESSOR_MIPS, "mips", NULL }, 1095 { PCI_CLASS_PROCESSOR_CO, "co-processor", NULL }, 1096 { 0xFF, NULL, NULL }, 1097 }; 1098 1099 static const PCIIFace usb_iface[] = { 1100 { PCI_CLASS_SERIAL_USB_UHCI, "usb-uhci" }, 1101 { PCI_CLASS_SERIAL_USB_OHCI, "usb-ohci", }, 1102 { PCI_CLASS_SERIAL_USB_EHCI, "usb-ehci" }, 1103 { PCI_CLASS_SERIAL_USB_XHCI, "usb-xhci" }, 1104 { PCI_CLASS_SERIAL_USB_UNKNOWN, "usb-unknown" }, 1105 { PCI_CLASS_SERIAL_USB_DEVICE, "usb-device" }, 1106 { 0xFF, NULL }, 1107 }; 1108 1109 static const PCISubClass ser_subclass[] = { 1110 { PCI_CLASS_SERIAL_FIREWIRE, "firewire", NULL }, 1111 { PCI_CLASS_SERIAL_ACCESS, "access-bus", NULL }, 1112 { PCI_CLASS_SERIAL_SSA, "ssa", NULL }, 1113 { PCI_CLASS_SERIAL_USB, "usb", usb_iface }, 1114 { PCI_CLASS_SERIAL_FIBER, "fibre-channel", NULL }, 1115 { PCI_CLASS_SERIAL_SMBUS, "smb", NULL }, 1116 { PCI_CLASS_SERIAL_IB, "infiniband", NULL }, 1117 { PCI_CLASS_SERIAL_IPMI, "ipmi", NULL }, 1118 { PCI_CLASS_SERIAL_SERCOS, "sercos", NULL }, 1119 { PCI_CLASS_SERIAL_CANBUS, "canbus", NULL }, 1120 { 0xFF, NULL, NULL }, 1121 }; 1122 1123 static const PCISubClass wrl_subclass[] = { 1124 { PCI_CLASS_WIRELESS_IRDA, "irda", NULL }, 1125 { PCI_CLASS_WIRELESS_CIR, "consumer-ir", NULL }, 1126 { PCI_CLASS_WIRELESS_RF_CONTROLLER, "rf-controller", NULL }, 1127 { PCI_CLASS_WIRELESS_BLUETOOTH, "bluetooth", NULL }, 1128 { PCI_CLASS_WIRELESS_BROADBAND, "broadband", NULL }, 1129 { 0xFF, NULL, NULL }, 1130 }; 1131 1132 static const PCISubClass sat_subclass[] = { 1133 { PCI_CLASS_SATELLITE_TV, "satellite-tv", NULL }, 1134 { PCI_CLASS_SATELLITE_AUDIO, "satellite-audio", NULL }, 1135 { PCI_CLASS_SATELLITE_VOICE, "satellite-voice", NULL }, 1136 { PCI_CLASS_SATELLITE_DATA, "satellite-data", NULL }, 1137 { 0xFF, NULL, NULL }, 1138 }; 1139 1140 static const PCISubClass crypt_subclass[] = { 1141 { PCI_CLASS_CRYPT_NETWORK, "network-encryption", NULL }, 1142 { PCI_CLASS_CRYPT_ENTERTAINMENT, 1143 "entertainment-encryption", NULL }, 1144 { 0xFF, NULL, NULL }, 1145 }; 1146 1147 static const PCISubClass spc_subclass[] = { 1148 { PCI_CLASS_SP_DPIO, "dpio", NULL }, 1149 { PCI_CLASS_SP_PERF, "counter", NULL }, 1150 { PCI_CLASS_SP_SYNCH, "measurement", NULL }, 1151 { PCI_CLASS_SP_MANAGEMENT, "management-card", NULL }, 1152 { 0xFF, NULL, NULL }, 1153 }; 1154 1155 static const PCIClass pci_classes[] = { 1156 { "legacy-device", undef_subclass }, 1157 { "mass-storage", mass_subclass }, 1158 { "network", net_subclass }, 1159 { "display", displ_subclass, }, 1160 { "multimedia-device", media_subclass }, 1161 { "memory-controller", mem_subclass }, 1162 { "unknown-bridge", bridg_subclass }, 1163 { "communication-controller", comm_subclass}, 1164 { "system-peripheral", sys_subclass }, 1165 { "input-controller", inp_subclass }, 1166 { "docking-station", dock_subclass }, 1167 { "cpu", cpu_subclass }, 1168 { "serial-bus", ser_subclass }, 1169 { "wireless-controller", wrl_subclass }, 1170 { "intelligent-io", NULL }, 1171 { "satellite-device", sat_subclass }, 1172 { "encryption", crypt_subclass }, 1173 { "data-processing-controller", spc_subclass }, 1174 }; 1175 1176 static const char *dt_name_from_class(uint8_t class, uint8_t subclass, 1177 uint8_t iface) 1178 { 1179 const PCIClass *pclass; 1180 const PCISubClass *psubclass; 1181 const PCIIFace *piface; 1182 const char *name; 1183 1184 if (class >= ARRAY_SIZE(pci_classes)) { 1185 return "pci"; 1186 } 1187 1188 pclass = pci_classes + class; 1189 name = pclass->name; 1190 1191 if (pclass->subc == NULL) { 1192 return name; 1193 } 1194 1195 psubclass = pclass->subc; 1196 while ((psubclass->subclass & 0xff) != 0xff) { 1197 if ((psubclass->subclass & 0xff) == subclass) { 1198 name = psubclass->name; 1199 break; 1200 } 1201 psubclass++; 1202 } 1203 1204 piface = psubclass->iface; 1205 if (piface == NULL) { 1206 return name; 1207 } 1208 while ((piface->iface & 0xff) != 0xff) { 1209 if ((piface->iface & 0xff) == iface) { 1210 name = piface->name; 1211 break; 1212 } 1213 piface++; 1214 } 1215 1216 return name; 1217 } 1218 1219 /* 1220 * DRC helper functions 1221 */ 1222 1223 static uint32_t drc_id_from_devfn(SpaprPhbState *phb, 1224 uint8_t chassis, int32_t devfn) 1225 { 1226 return (phb->index << 16) | (chassis << 8) | devfn; 1227 } 1228 1229 static SpaprDrc *drc_from_devfn(SpaprPhbState *phb, 1230 uint8_t chassis, int32_t devfn) 1231 { 1232 return spapr_drc_by_id(TYPE_SPAPR_DRC_PCI, 1233 drc_id_from_devfn(phb, chassis, devfn)); 1234 } 1235 1236 static uint8_t chassis_from_bus(PCIBus *bus, Error **errp) 1237 { 1238 if (pci_bus_is_root(bus)) { 1239 return 0; 1240 } else { 1241 PCIDevice *bridge = pci_bridge_get_device(bus); 1242 1243 return object_property_get_uint(OBJECT(bridge), "chassis_nr", errp); 1244 } 1245 } 1246 1247 static SpaprDrc *drc_from_dev(SpaprPhbState *phb, PCIDevice *dev) 1248 { 1249 Error *local_err = NULL; 1250 uint8_t chassis = chassis_from_bus(pci_get_bus(dev), &local_err); 1251 1252 if (local_err) { 1253 error_report_err(local_err); 1254 return NULL; 1255 } 1256 1257 return drc_from_devfn(phb, chassis, dev->devfn); 1258 } 1259 1260 static void add_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp) 1261 { 1262 Object *owner; 1263 int i; 1264 uint8_t chassis; 1265 Error *local_err = NULL; 1266 1267 if (!phb->dr_enabled) { 1268 return; 1269 } 1270 1271 chassis = chassis_from_bus(bus, &local_err); 1272 if (local_err) { 1273 error_propagate(errp, local_err); 1274 return; 1275 } 1276 1277 if (pci_bus_is_root(bus)) { 1278 owner = OBJECT(phb); 1279 } else { 1280 owner = OBJECT(pci_bridge_get_device(bus)); 1281 } 1282 1283 for (i = 0; i < PCI_SLOT_MAX * PCI_FUNC_MAX; i++) { 1284 spapr_dr_connector_new(owner, TYPE_SPAPR_DRC_PCI, 1285 drc_id_from_devfn(phb, chassis, i)); 1286 } 1287 } 1288 1289 static void remove_drcs(SpaprPhbState *phb, PCIBus *bus, Error **errp) 1290 { 1291 int i; 1292 uint8_t chassis; 1293 Error *local_err = NULL; 1294 1295 if (!phb->dr_enabled) { 1296 return; 1297 } 1298 1299 chassis = chassis_from_bus(bus, &local_err); 1300 if (local_err) { 1301 error_propagate(errp, local_err); 1302 return; 1303 } 1304 1305 for (i = PCI_SLOT_MAX * PCI_FUNC_MAX - 1; i >= 0; i--) { 1306 SpaprDrc *drc = drc_from_devfn(phb, chassis, i); 1307 1308 if (drc) { 1309 object_unparent(OBJECT(drc)); 1310 } 1311 } 1312 } 1313 1314 typedef struct PciWalkFdt { 1315 void *fdt; 1316 int offset; 1317 SpaprPhbState *sphb; 1318 int err; 1319 } PciWalkFdt; 1320 1321 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev, 1322 void *fdt, int parent_offset); 1323 1324 static void spapr_dt_pci_device_cb(PCIBus *bus, PCIDevice *pdev, 1325 void *opaque) 1326 { 1327 PciWalkFdt *p = opaque; 1328 int err; 1329 1330 if (p->err) { 1331 /* Something's already broken, don't keep going */ 1332 return; 1333 } 1334 1335 err = spapr_dt_pci_device(p->sphb, pdev, p->fdt, p->offset); 1336 if (err < 0) { 1337 p->err = err; 1338 } 1339 } 1340 1341 /* Augment PCI device node with bridge specific information */ 1342 static int spapr_dt_pci_bus(SpaprPhbState *sphb, PCIBus *bus, 1343 void *fdt, int offset) 1344 { 1345 PciWalkFdt cbinfo = { 1346 .fdt = fdt, 1347 .offset = offset, 1348 .sphb = sphb, 1349 .err = 0, 1350 }; 1351 int ret; 1352 1353 _FDT(fdt_setprop_cell(fdt, offset, "#address-cells", 1354 RESOURCE_CELLS_ADDRESS)); 1355 _FDT(fdt_setprop_cell(fdt, offset, "#size-cells", 1356 RESOURCE_CELLS_SIZE)); 1357 1358 if (bus) { 1359 pci_for_each_device_reverse(bus, pci_bus_num(bus), 1360 spapr_dt_pci_device_cb, &cbinfo); 1361 if (cbinfo.err) { 1362 return cbinfo.err; 1363 } 1364 } 1365 1366 ret = spapr_dt_drc(fdt, offset, OBJECT(bus->parent_dev), 1367 SPAPR_DR_CONNECTOR_TYPE_PCI); 1368 if (ret) { 1369 return ret; 1370 } 1371 1372 return offset; 1373 } 1374 1375 /* create OF node for pci device and required OF DT properties */ 1376 static int spapr_dt_pci_device(SpaprPhbState *sphb, PCIDevice *dev, 1377 void *fdt, int parent_offset) 1378 { 1379 int offset; 1380 const gchar *basename; 1381 gchar *nodename; 1382 int slot = PCI_SLOT(dev->devfn); 1383 int func = PCI_FUNC(dev->devfn); 1384 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(dev); 1385 ResourceProps rp; 1386 SpaprDrc *drc = drc_from_dev(sphb, dev); 1387 uint32_t vendor_id = pci_default_read_config(dev, PCI_VENDOR_ID, 2); 1388 uint32_t device_id = pci_default_read_config(dev, PCI_DEVICE_ID, 2); 1389 uint32_t revision_id = pci_default_read_config(dev, PCI_REVISION_ID, 1); 1390 uint32_t ccode = pci_default_read_config(dev, PCI_CLASS_PROG, 3); 1391 uint32_t irq_pin = pci_default_read_config(dev, PCI_INTERRUPT_PIN, 1); 1392 uint32_t subsystem_id = pci_default_read_config(dev, PCI_SUBSYSTEM_ID, 2); 1393 uint32_t subsystem_vendor_id = 1394 pci_default_read_config(dev, PCI_SUBSYSTEM_VENDOR_ID, 2); 1395 uint32_t cache_line_size = 1396 pci_default_read_config(dev, PCI_CACHE_LINE_SIZE, 1); 1397 uint32_t pci_status = pci_default_read_config(dev, PCI_STATUS, 2); 1398 gchar *loc_code; 1399 1400 basename = dt_name_from_class((ccode >> 16) & 0xff, (ccode >> 8) & 0xff, 1401 ccode & 0xff); 1402 1403 if (func != 0) { 1404 nodename = g_strdup_printf("%s@%x,%x", basename, slot, func); 1405 } else { 1406 nodename = g_strdup_printf("%s@%x", basename, slot); 1407 } 1408 1409 _FDT(offset = fdt_add_subnode(fdt, parent_offset, nodename)); 1410 1411 g_free(nodename); 1412 1413 /* in accordance with PAPR+ v2.7 13.6.3, Table 181 */ 1414 _FDT(fdt_setprop_cell(fdt, offset, "vendor-id", vendor_id)); 1415 _FDT(fdt_setprop_cell(fdt, offset, "device-id", device_id)); 1416 _FDT(fdt_setprop_cell(fdt, offset, "revision-id", revision_id)); 1417 1418 _FDT(fdt_setprop_cell(fdt, offset, "class-code", ccode)); 1419 if (irq_pin) { 1420 _FDT(fdt_setprop_cell(fdt, offset, "interrupts", irq_pin)); 1421 } 1422 1423 if (subsystem_id) { 1424 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-id", subsystem_id)); 1425 } 1426 1427 if (subsystem_vendor_id) { 1428 _FDT(fdt_setprop_cell(fdt, offset, "subsystem-vendor-id", 1429 subsystem_vendor_id)); 1430 } 1431 1432 _FDT(fdt_setprop_cell(fdt, offset, "cache-line-size", cache_line_size)); 1433 1434 1435 /* the following fdt cells are masked off the pci status register */ 1436 _FDT(fdt_setprop_cell(fdt, offset, "devsel-speed", 1437 PCI_STATUS_DEVSEL_MASK & pci_status)); 1438 1439 if (pci_status & PCI_STATUS_FAST_BACK) { 1440 _FDT(fdt_setprop(fdt, offset, "fast-back-to-back", NULL, 0)); 1441 } 1442 if (pci_status & PCI_STATUS_66MHZ) { 1443 _FDT(fdt_setprop(fdt, offset, "66mhz-capable", NULL, 0)); 1444 } 1445 if (pci_status & PCI_STATUS_UDF) { 1446 _FDT(fdt_setprop(fdt, offset, "udf-supported", NULL, 0)); 1447 } 1448 1449 loc_code = spapr_phb_get_loc_code(sphb, dev); 1450 _FDT(fdt_setprop_string(fdt, offset, "ibm,loc-code", loc_code)); 1451 g_free(loc_code); 1452 1453 if (drc) { 1454 _FDT(fdt_setprop_cell(fdt, offset, "ibm,my-drc-index", 1455 spapr_drc_index(drc))); 1456 } 1457 1458 if (msi_present(dev)) { 1459 uint32_t max_msi = msi_nr_vectors_allocated(dev); 1460 if (max_msi) { 1461 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi", max_msi)); 1462 } 1463 } 1464 if (msix_present(dev)) { 1465 uint32_t max_msix = dev->msix_entries_nr; 1466 if (max_msix) { 1467 _FDT(fdt_setprop_cell(fdt, offset, "ibm,req#msi-x", max_msix)); 1468 } 1469 } 1470 1471 populate_resource_props(dev, &rp); 1472 _FDT(fdt_setprop(fdt, offset, "reg", (uint8_t *)rp.reg, rp.reg_len)); 1473 _FDT(fdt_setprop(fdt, offset, "assigned-addresses", 1474 (uint8_t *)rp.assigned, rp.assigned_len)); 1475 1476 if (sphb->pcie_ecs && pci_is_express(dev)) { 1477 _FDT(fdt_setprop_cell(fdt, offset, "ibm,pci-config-space-type", 0x1)); 1478 } 1479 1480 spapr_phb_nvgpu_populate_pcidev_dt(dev, fdt, offset, sphb); 1481 1482 if (!pc->is_bridge) { 1483 /* Properties only for non-bridges */ 1484 uint32_t min_grant = pci_default_read_config(dev, PCI_MIN_GNT, 1); 1485 uint32_t max_latency = pci_default_read_config(dev, PCI_MAX_LAT, 1); 1486 _FDT(fdt_setprop_cell(fdt, offset, "min-grant", min_grant)); 1487 _FDT(fdt_setprop_cell(fdt, offset, "max-latency", max_latency)); 1488 return offset; 1489 } else { 1490 PCIBus *sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(dev)); 1491 1492 return spapr_dt_pci_bus(sphb, sec_bus, fdt, offset); 1493 } 1494 } 1495 1496 /* Callback to be called during DRC release. */ 1497 void spapr_phb_remove_pci_device_cb(DeviceState *dev) 1498 { 1499 HotplugHandler *hotplug_ctrl = qdev_get_hotplug_handler(dev); 1500 1501 hotplug_handler_unplug(hotplug_ctrl, dev, &error_abort); 1502 object_unparent(OBJECT(dev)); 1503 } 1504 1505 int spapr_pci_dt_populate(SpaprDrc *drc, SpaprMachineState *spapr, 1506 void *fdt, int *fdt_start_offset, Error **errp) 1507 { 1508 HotplugHandler *plug_handler = qdev_get_hotplug_handler(drc->dev); 1509 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(plug_handler); 1510 PCIDevice *pdev = PCI_DEVICE(drc->dev); 1511 1512 *fdt_start_offset = spapr_dt_pci_device(sphb, pdev, fdt, 0); 1513 return 0; 1514 } 1515 1516 static void spapr_pci_bridge_plug(SpaprPhbState *phb, 1517 PCIBridge *bridge, 1518 Error **errp) 1519 { 1520 Error *local_err = NULL; 1521 PCIBus *bus = pci_bridge_get_sec_bus(bridge); 1522 1523 add_drcs(phb, bus, &local_err); 1524 if (local_err) { 1525 error_propagate(errp, local_err); 1526 return; 1527 } 1528 } 1529 1530 static void spapr_pci_plug(HotplugHandler *plug_handler, 1531 DeviceState *plugged_dev, Error **errp) 1532 { 1533 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1534 PCIDevice *pdev = PCI_DEVICE(plugged_dev); 1535 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev); 1536 SpaprDrc *drc = drc_from_dev(phb, pdev); 1537 Error *local_err = NULL; 1538 PCIBus *bus = PCI_BUS(qdev_get_parent_bus(DEVICE(pdev))); 1539 uint32_t slotnr = PCI_SLOT(pdev->devfn); 1540 1541 /* if DR is disabled we don't need to do anything in the case of 1542 * hotplug or coldplug callbacks 1543 */ 1544 if (!phb->dr_enabled) { 1545 /* if this is a hotplug operation initiated by the user 1546 * we need to let them know it's not enabled 1547 */ 1548 if (plugged_dev->hotplugged) { 1549 error_setg(&local_err, QERR_BUS_NO_HOTPLUG, 1550 object_get_typename(OBJECT(phb))); 1551 } 1552 goto out; 1553 } 1554 1555 g_assert(drc); 1556 1557 if (pc->is_bridge) { 1558 spapr_pci_bridge_plug(phb, PCI_BRIDGE(plugged_dev), &local_err); 1559 if (local_err) { 1560 error_propagate(errp, local_err); 1561 return; 1562 } 1563 } 1564 1565 /* Following the QEMU convention used for PCIe multifunction 1566 * hotplug, we do not allow functions to be hotplugged to a 1567 * slot that already has function 0 present 1568 */ 1569 if (plugged_dev->hotplugged && bus->devices[PCI_DEVFN(slotnr, 0)] && 1570 PCI_FUNC(pdev->devfn) != 0) { 1571 error_setg(&local_err, "PCI: slot %d function 0 already ocuppied by %s," 1572 " additional functions can no longer be exposed to guest.", 1573 slotnr, bus->devices[PCI_DEVFN(slotnr, 0)]->name); 1574 goto out; 1575 } 1576 1577 spapr_drc_attach(drc, DEVICE(pdev), &local_err); 1578 if (local_err) { 1579 goto out; 1580 } 1581 1582 /* If this is function 0, signal hotplug for all the device functions. 1583 * Otherwise defer sending the hotplug event. 1584 */ 1585 if (!spapr_drc_hotplugged(plugged_dev)) { 1586 spapr_drc_reset(drc); 1587 } else if (PCI_FUNC(pdev->devfn) == 0) { 1588 int i; 1589 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err); 1590 1591 if (local_err) { 1592 error_propagate(errp, local_err); 1593 return; 1594 } 1595 1596 for (i = 0; i < 8; i++) { 1597 SpaprDrc *func_drc; 1598 SpaprDrcClass *func_drck; 1599 SpaprDREntitySense state; 1600 1601 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1602 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1603 state = func_drck->dr_entity_sense(func_drc); 1604 1605 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) { 1606 spapr_hotplug_req_add_by_index(func_drc); 1607 } 1608 } 1609 } 1610 1611 out: 1612 error_propagate(errp, local_err); 1613 } 1614 1615 static void spapr_pci_bridge_unplug(SpaprPhbState *phb, 1616 PCIBridge *bridge, 1617 Error **errp) 1618 { 1619 Error *local_err = NULL; 1620 PCIBus *bus = pci_bridge_get_sec_bus(bridge); 1621 1622 remove_drcs(phb, bus, &local_err); 1623 if (local_err) { 1624 error_propagate(errp, local_err); 1625 return; 1626 } 1627 } 1628 1629 static void spapr_pci_unplug(HotplugHandler *plug_handler, 1630 DeviceState *plugged_dev, Error **errp) 1631 { 1632 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev); 1633 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1634 1635 /* some version guests do not wait for completion of a device 1636 * cleanup (generally done asynchronously by the kernel) before 1637 * signaling to QEMU that the device is safe, but instead sleep 1638 * for some 'safe' period of time. unfortunately on a busy host 1639 * this sleep isn't guaranteed to be long enough, resulting in 1640 * bad things like IRQ lines being left asserted during final 1641 * device removal. to deal with this we call reset just prior 1642 * to finalizing the device, which will put the device back into 1643 * an 'idle' state, as the device cleanup code expects. 1644 */ 1645 pci_device_reset(PCI_DEVICE(plugged_dev)); 1646 1647 if (pc->is_bridge) { 1648 Error *local_err = NULL; 1649 spapr_pci_bridge_unplug(phb, PCI_BRIDGE(plugged_dev), &local_err); 1650 if (local_err) { 1651 error_propagate(errp, local_err); 1652 } 1653 return; 1654 } 1655 1656 object_property_set_bool(OBJECT(plugged_dev), false, "realized", NULL); 1657 } 1658 1659 static void spapr_pci_unplug_request(HotplugHandler *plug_handler, 1660 DeviceState *plugged_dev, Error **errp) 1661 { 1662 SpaprPhbState *phb = SPAPR_PCI_HOST_BRIDGE(DEVICE(plug_handler)); 1663 PCIDevice *pdev = PCI_DEVICE(plugged_dev); 1664 SpaprDrc *drc = drc_from_dev(phb, pdev); 1665 1666 if (!phb->dr_enabled) { 1667 error_setg(errp, QERR_BUS_NO_HOTPLUG, 1668 object_get_typename(OBJECT(phb))); 1669 return; 1670 } 1671 1672 g_assert(drc); 1673 g_assert(drc->dev == plugged_dev); 1674 1675 if (!spapr_drc_unplug_requested(drc)) { 1676 PCIDeviceClass *pc = PCI_DEVICE_GET_CLASS(plugged_dev); 1677 uint32_t slotnr = PCI_SLOT(pdev->devfn); 1678 SpaprDrc *func_drc; 1679 SpaprDrcClass *func_drck; 1680 SpaprDREntitySense state; 1681 int i; 1682 Error *local_err = NULL; 1683 uint8_t chassis = chassis_from_bus(pci_get_bus(pdev), &local_err); 1684 1685 if (local_err) { 1686 error_propagate(errp, local_err); 1687 return; 1688 } 1689 1690 if (pc->is_bridge) { 1691 error_setg(errp, "PCI: Hot unplug of PCI bridges not supported"); 1692 } 1693 1694 /* ensure any other present functions are pending unplug */ 1695 if (PCI_FUNC(pdev->devfn) == 0) { 1696 for (i = 1; i < 8; i++) { 1697 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1698 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1699 state = func_drck->dr_entity_sense(func_drc); 1700 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT 1701 && !spapr_drc_unplug_requested(func_drc)) { 1702 error_setg(errp, 1703 "PCI: slot %d, function %d still present. " 1704 "Must unplug all non-0 functions first.", 1705 slotnr, i); 1706 return; 1707 } 1708 } 1709 } 1710 1711 spapr_drc_detach(drc); 1712 1713 /* if this isn't func 0, defer unplug event. otherwise signal removal 1714 * for all present functions 1715 */ 1716 if (PCI_FUNC(pdev->devfn) == 0) { 1717 for (i = 7; i >= 0; i--) { 1718 func_drc = drc_from_devfn(phb, chassis, PCI_DEVFN(slotnr, i)); 1719 func_drck = SPAPR_DR_CONNECTOR_GET_CLASS(func_drc); 1720 state = func_drck->dr_entity_sense(func_drc); 1721 if (state == SPAPR_DR_ENTITY_SENSE_PRESENT) { 1722 spapr_hotplug_req_remove_by_index(func_drc); 1723 } 1724 } 1725 } 1726 } 1727 } 1728 1729 static void spapr_phb_finalizefn(Object *obj) 1730 { 1731 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(obj); 1732 1733 g_free(sphb->dtbusname); 1734 sphb->dtbusname = NULL; 1735 } 1736 1737 static void spapr_phb_unrealize(DeviceState *dev, Error **errp) 1738 { 1739 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 1740 SysBusDevice *s = SYS_BUS_DEVICE(dev); 1741 PCIHostState *phb = PCI_HOST_BRIDGE(s); 1742 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(phb); 1743 SpaprTceTable *tcet; 1744 int i; 1745 const unsigned windows_supported = spapr_phb_windows_supported(sphb); 1746 Error *local_err = NULL; 1747 1748 spapr_phb_nvgpu_free(sphb); 1749 1750 if (sphb->msi) { 1751 g_hash_table_unref(sphb->msi); 1752 sphb->msi = NULL; 1753 } 1754 1755 /* 1756 * Remove IO/MMIO subregions and aliases, rest should get cleaned 1757 * via PHB's unrealize->object_finalize 1758 */ 1759 for (i = windows_supported - 1; i >= 0; i--) { 1760 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]); 1761 if (tcet) { 1762 memory_region_del_subregion(&sphb->iommu_root, 1763 spapr_tce_get_iommu(tcet)); 1764 } 1765 } 1766 1767 remove_drcs(sphb, phb->bus, &local_err); 1768 if (local_err) { 1769 error_propagate(errp, local_err); 1770 return; 1771 } 1772 1773 for (i = PCI_NUM_PINS - 1; i >= 0; i--) { 1774 if (sphb->lsi_table[i].irq) { 1775 spapr_irq_free(spapr, sphb->lsi_table[i].irq, 1); 1776 sphb->lsi_table[i].irq = 0; 1777 } 1778 } 1779 1780 QLIST_REMOVE(sphb, list); 1781 1782 memory_region_del_subregion(&sphb->iommu_root, &sphb->msiwindow); 1783 1784 address_space_destroy(&sphb->iommu_as); 1785 1786 qbus_set_hotplug_handler(BUS(phb->bus), NULL, &error_abort); 1787 pci_unregister_root_bus(phb->bus); 1788 1789 memory_region_del_subregion(get_system_memory(), &sphb->iowindow); 1790 if (sphb->mem64_win_pciaddr != (hwaddr)-1) { 1791 memory_region_del_subregion(get_system_memory(), &sphb->mem64window); 1792 } 1793 memory_region_del_subregion(get_system_memory(), &sphb->mem32window); 1794 } 1795 1796 static void spapr_phb_realize(DeviceState *dev, Error **errp) 1797 { 1798 /* We don't use SPAPR_MACHINE() in order to exit gracefully if the user 1799 * tries to add a sPAPR PHB to a non-pseries machine. 1800 */ 1801 SpaprMachineState *spapr = 1802 (SpaprMachineState *) object_dynamic_cast(qdev_get_machine(), 1803 TYPE_SPAPR_MACHINE); 1804 SpaprMachineClass *smc = spapr ? SPAPR_MACHINE_GET_CLASS(spapr) : NULL; 1805 SysBusDevice *s = SYS_BUS_DEVICE(dev); 1806 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(s); 1807 PCIHostState *phb = PCI_HOST_BRIDGE(s); 1808 char *namebuf; 1809 int i; 1810 PCIBus *bus; 1811 uint64_t msi_window_size = 4096; 1812 SpaprTceTable *tcet; 1813 const unsigned windows_supported = spapr_phb_windows_supported(sphb); 1814 Error *local_err = NULL; 1815 1816 if (!spapr) { 1817 error_setg(errp, TYPE_SPAPR_PCI_HOST_BRIDGE " needs a pseries machine"); 1818 return; 1819 } 1820 1821 assert(sphb->index != (uint32_t)-1); /* checked in spapr_phb_pre_plug() */ 1822 1823 if (sphb->mem64_win_size != 0) { 1824 if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) { 1825 error_setg(errp, "32-bit memory window of size 0x%"HWADDR_PRIx 1826 " (max 2 GiB)", sphb->mem_win_size); 1827 return; 1828 } 1829 1830 /* 64-bit window defaults to identity mapping */ 1831 sphb->mem64_win_pciaddr = sphb->mem64_win_addr; 1832 } else if (sphb->mem_win_size > SPAPR_PCI_MEM32_WIN_SIZE) { 1833 /* 1834 * For compatibility with old configuration, if no 64-bit MMIO 1835 * window is specified, but the ordinary (32-bit) memory 1836 * window is specified as > 2GiB, we treat it as a 2GiB 32-bit 1837 * window, with a 64-bit MMIO window following on immediately 1838 * afterwards 1839 */ 1840 sphb->mem64_win_size = sphb->mem_win_size - SPAPR_PCI_MEM32_WIN_SIZE; 1841 sphb->mem64_win_addr = sphb->mem_win_addr + SPAPR_PCI_MEM32_WIN_SIZE; 1842 sphb->mem64_win_pciaddr = 1843 SPAPR_PCI_MEM_WIN_BUS_OFFSET + SPAPR_PCI_MEM32_WIN_SIZE; 1844 sphb->mem_win_size = SPAPR_PCI_MEM32_WIN_SIZE; 1845 } 1846 1847 if (spapr_pci_find_phb(spapr, sphb->buid)) { 1848 SpaprPhbState *s; 1849 1850 error_setg(errp, "PCI host bridges must have unique indexes"); 1851 error_append_hint(errp, "The following indexes are already in use:"); 1852 QLIST_FOREACH(s, &spapr->phbs, list) { 1853 error_append_hint(errp, " %d", s->index); 1854 } 1855 error_append_hint(errp, "\nTry another value for the index property\n"); 1856 return; 1857 } 1858 1859 if (sphb->numa_node != -1 && 1860 (sphb->numa_node >= MAX_NODES || !numa_info[sphb->numa_node].present)) { 1861 error_setg(errp, "Invalid NUMA node ID for PCI host bridge"); 1862 return; 1863 } 1864 1865 sphb->dtbusname = g_strdup_printf("pci@%" PRIx64, sphb->buid); 1866 1867 /* Initialize memory regions */ 1868 namebuf = g_strdup_printf("%s.mmio", sphb->dtbusname); 1869 memory_region_init(&sphb->memspace, OBJECT(sphb), namebuf, UINT64_MAX); 1870 g_free(namebuf); 1871 1872 namebuf = g_strdup_printf("%s.mmio32-alias", sphb->dtbusname); 1873 memory_region_init_alias(&sphb->mem32window, OBJECT(sphb), 1874 namebuf, &sphb->memspace, 1875 SPAPR_PCI_MEM_WIN_BUS_OFFSET, sphb->mem_win_size); 1876 g_free(namebuf); 1877 memory_region_add_subregion(get_system_memory(), sphb->mem_win_addr, 1878 &sphb->mem32window); 1879 1880 if (sphb->mem64_win_size != 0) { 1881 namebuf = g_strdup_printf("%s.mmio64-alias", sphb->dtbusname); 1882 memory_region_init_alias(&sphb->mem64window, OBJECT(sphb), 1883 namebuf, &sphb->memspace, 1884 sphb->mem64_win_pciaddr, sphb->mem64_win_size); 1885 g_free(namebuf); 1886 1887 memory_region_add_subregion(get_system_memory(), 1888 sphb->mem64_win_addr, 1889 &sphb->mem64window); 1890 } 1891 1892 /* Initialize IO regions */ 1893 namebuf = g_strdup_printf("%s.io", sphb->dtbusname); 1894 memory_region_init(&sphb->iospace, OBJECT(sphb), 1895 namebuf, SPAPR_PCI_IO_WIN_SIZE); 1896 g_free(namebuf); 1897 1898 namebuf = g_strdup_printf("%s.io-alias", sphb->dtbusname); 1899 memory_region_init_alias(&sphb->iowindow, OBJECT(sphb), namebuf, 1900 &sphb->iospace, 0, SPAPR_PCI_IO_WIN_SIZE); 1901 g_free(namebuf); 1902 memory_region_add_subregion(get_system_memory(), sphb->io_win_addr, 1903 &sphb->iowindow); 1904 1905 bus = pci_register_root_bus(dev, NULL, 1906 pci_spapr_set_irq, pci_swizzle_map_irq_fn, sphb, 1907 &sphb->memspace, &sphb->iospace, 1908 PCI_DEVFN(0, 0), PCI_NUM_PINS, 1909 TYPE_PCI_BUS); 1910 1911 /* 1912 * Despite resembling a vanilla PCI bus in most ways, the PAPR 1913 * para-virtualized PCI bus *does* permit PCI-E extended config 1914 * space access 1915 */ 1916 if (sphb->pcie_ecs) { 1917 bus->flags |= PCI_BUS_EXTENDED_CONFIG_SPACE; 1918 } 1919 phb->bus = bus; 1920 qbus_set_hotplug_handler(BUS(phb->bus), OBJECT(sphb), NULL); 1921 1922 /* 1923 * Initialize PHB address space. 1924 * By default there will be at least one subregion for default 1925 * 32bit DMA window. 1926 * Later the guest might want to create another DMA window 1927 * which will become another memory subregion. 1928 */ 1929 namebuf = g_strdup_printf("%s.iommu-root", sphb->dtbusname); 1930 memory_region_init(&sphb->iommu_root, OBJECT(sphb), 1931 namebuf, UINT64_MAX); 1932 g_free(namebuf); 1933 address_space_init(&sphb->iommu_as, &sphb->iommu_root, 1934 sphb->dtbusname); 1935 1936 /* 1937 * As MSI/MSIX interrupts trigger by writing at MSI/MSIX vectors, 1938 * we need to allocate some memory to catch those writes coming 1939 * from msi_notify()/msix_notify(). 1940 * As MSIMessage:addr is going to be the same and MSIMessage:data 1941 * is going to be a VIRQ number, 4 bytes of the MSI MR will only 1942 * be used. 1943 * 1944 * For KVM we want to ensure that this memory is a full page so that 1945 * our memory slot is of page size granularity. 1946 */ 1947 #ifdef CONFIG_KVM 1948 if (kvm_enabled()) { 1949 msi_window_size = getpagesize(); 1950 } 1951 #endif 1952 1953 memory_region_init_io(&sphb->msiwindow, OBJECT(sphb), &spapr_msi_ops, spapr, 1954 "msi", msi_window_size); 1955 memory_region_add_subregion(&sphb->iommu_root, SPAPR_PCI_MSI_WINDOW, 1956 &sphb->msiwindow); 1957 1958 pci_setup_iommu(bus, spapr_pci_dma_iommu, sphb); 1959 1960 pci_bus_set_route_irq_fn(bus, spapr_route_intx_pin_to_irq); 1961 1962 QLIST_INSERT_HEAD(&spapr->phbs, sphb, list); 1963 1964 /* Initialize the LSI table */ 1965 for (i = 0; i < PCI_NUM_PINS; i++) { 1966 uint32_t irq = SPAPR_IRQ_PCI_LSI + sphb->index * PCI_NUM_PINS + i; 1967 1968 if (smc->legacy_irq_allocation) { 1969 irq = spapr_irq_findone(spapr, &local_err); 1970 if (local_err) { 1971 error_propagate_prepend(errp, local_err, 1972 "can't allocate LSIs: "); 1973 /* 1974 * Older machines will never support PHB hotplug, ie, this is an 1975 * init only path and QEMU will terminate. No need to rollback. 1976 */ 1977 return; 1978 } 1979 } 1980 1981 spapr_irq_claim(spapr, irq, true, &local_err); 1982 if (local_err) { 1983 error_propagate_prepend(errp, local_err, "can't allocate LSIs: "); 1984 goto unrealize; 1985 } 1986 1987 sphb->lsi_table[i].irq = irq; 1988 } 1989 1990 /* allocate connectors for child PCI devices */ 1991 add_drcs(sphb, phb->bus, &local_err); 1992 if (local_err) { 1993 error_propagate(errp, local_err); 1994 goto unrealize; 1995 } 1996 1997 /* DMA setup */ 1998 for (i = 0; i < windows_supported; ++i) { 1999 tcet = spapr_tce_new_table(DEVICE(sphb), sphb->dma_liobn[i]); 2000 if (!tcet) { 2001 error_setg(errp, "Creating window#%d failed for %s", 2002 i, sphb->dtbusname); 2003 goto unrealize; 2004 } 2005 memory_region_add_subregion(&sphb->iommu_root, 0, 2006 spapr_tce_get_iommu(tcet)); 2007 } 2008 2009 sphb->msi = g_hash_table_new_full(g_int_hash, g_int_equal, g_free, g_free); 2010 return; 2011 2012 unrealize: 2013 spapr_phb_unrealize(dev, NULL); 2014 } 2015 2016 static int spapr_phb_children_reset(Object *child, void *opaque) 2017 { 2018 DeviceState *dev = (DeviceState *) object_dynamic_cast(child, TYPE_DEVICE); 2019 2020 if (dev) { 2021 device_reset(dev); 2022 } 2023 2024 return 0; 2025 } 2026 2027 void spapr_phb_dma_reset(SpaprPhbState *sphb) 2028 { 2029 int i; 2030 SpaprTceTable *tcet; 2031 2032 for (i = 0; i < SPAPR_PCI_DMA_MAX_WINDOWS; ++i) { 2033 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[i]); 2034 2035 if (tcet && tcet->nb_table) { 2036 spapr_tce_table_disable(tcet); 2037 } 2038 } 2039 2040 /* Register default 32bit DMA window */ 2041 tcet = spapr_tce_find_by_liobn(sphb->dma_liobn[0]); 2042 spapr_tce_table_enable(tcet, SPAPR_TCE_PAGE_SHIFT, sphb->dma_win_addr, 2043 sphb->dma_win_size >> SPAPR_TCE_PAGE_SHIFT); 2044 } 2045 2046 static void spapr_phb_reset(DeviceState *qdev) 2047 { 2048 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(qdev); 2049 Error *errp = NULL; 2050 2051 spapr_phb_dma_reset(sphb); 2052 spapr_phb_nvgpu_free(sphb); 2053 spapr_phb_nvgpu_setup(sphb, &errp); 2054 if (errp) { 2055 error_report_err(errp); 2056 } 2057 2058 /* Reset the IOMMU state */ 2059 object_child_foreach(OBJECT(qdev), spapr_phb_children_reset, NULL); 2060 2061 if (spapr_phb_eeh_available(SPAPR_PCI_HOST_BRIDGE(qdev))) { 2062 spapr_phb_vfio_reset(qdev); 2063 } 2064 } 2065 2066 static Property spapr_phb_properties[] = { 2067 DEFINE_PROP_UINT32("index", SpaprPhbState, index, -1), 2068 DEFINE_PROP_UINT64("mem_win_size", SpaprPhbState, mem_win_size, 2069 SPAPR_PCI_MEM32_WIN_SIZE), 2070 DEFINE_PROP_UINT64("mem64_win_size", SpaprPhbState, mem64_win_size, 2071 SPAPR_PCI_MEM64_WIN_SIZE), 2072 DEFINE_PROP_UINT64("io_win_size", SpaprPhbState, io_win_size, 2073 SPAPR_PCI_IO_WIN_SIZE), 2074 DEFINE_PROP_BOOL("dynamic-reconfiguration", SpaprPhbState, dr_enabled, 2075 true), 2076 /* Default DMA window is 0..1GB */ 2077 DEFINE_PROP_UINT64("dma_win_addr", SpaprPhbState, dma_win_addr, 0), 2078 DEFINE_PROP_UINT64("dma_win_size", SpaprPhbState, dma_win_size, 0x40000000), 2079 DEFINE_PROP_UINT64("dma64_win_addr", SpaprPhbState, dma64_win_addr, 2080 0x800000000000000ULL), 2081 DEFINE_PROP_BOOL("ddw", SpaprPhbState, ddw_enabled, true), 2082 DEFINE_PROP_UINT64("pgsz", SpaprPhbState, page_size_mask, 2083 (1ULL << 12) | (1ULL << 16)), 2084 DEFINE_PROP_UINT32("numa_node", SpaprPhbState, numa_node, -1), 2085 DEFINE_PROP_BOOL("pre-2.8-migration", SpaprPhbState, 2086 pre_2_8_migration, false), 2087 DEFINE_PROP_BOOL("pcie-extended-configuration-space", SpaprPhbState, 2088 pcie_ecs, true), 2089 DEFINE_PROP_UINT64("gpa", SpaprPhbState, nv2_gpa_win_addr, 0), 2090 DEFINE_PROP_UINT64("atsd", SpaprPhbState, nv2_atsd_win_addr, 0), 2091 DEFINE_PROP_END_OF_LIST(), 2092 }; 2093 2094 static const VMStateDescription vmstate_spapr_pci_lsi = { 2095 .name = "spapr_pci/lsi", 2096 .version_id = 1, 2097 .minimum_version_id = 1, 2098 .fields = (VMStateField[]) { 2099 VMSTATE_UINT32_EQUAL(irq, struct spapr_pci_lsi, NULL), 2100 2101 VMSTATE_END_OF_LIST() 2102 }, 2103 }; 2104 2105 static const VMStateDescription vmstate_spapr_pci_msi = { 2106 .name = "spapr_pci/msi", 2107 .version_id = 1, 2108 .minimum_version_id = 1, 2109 .fields = (VMStateField []) { 2110 VMSTATE_UINT32(key, spapr_pci_msi_mig), 2111 VMSTATE_UINT32(value.first_irq, spapr_pci_msi_mig), 2112 VMSTATE_UINT32(value.num, spapr_pci_msi_mig), 2113 VMSTATE_END_OF_LIST() 2114 }, 2115 }; 2116 2117 static int spapr_pci_pre_save(void *opaque) 2118 { 2119 SpaprPhbState *sphb = opaque; 2120 GHashTableIter iter; 2121 gpointer key, value; 2122 int i; 2123 2124 if (sphb->pre_2_8_migration) { 2125 sphb->mig_liobn = sphb->dma_liobn[0]; 2126 sphb->mig_mem_win_addr = sphb->mem_win_addr; 2127 sphb->mig_mem_win_size = sphb->mem_win_size; 2128 sphb->mig_io_win_addr = sphb->io_win_addr; 2129 sphb->mig_io_win_size = sphb->io_win_size; 2130 2131 if ((sphb->mem64_win_size != 0) 2132 && (sphb->mem64_win_addr 2133 == (sphb->mem_win_addr + sphb->mem_win_size))) { 2134 sphb->mig_mem_win_size += sphb->mem64_win_size; 2135 } 2136 } 2137 2138 g_free(sphb->msi_devs); 2139 sphb->msi_devs = NULL; 2140 sphb->msi_devs_num = g_hash_table_size(sphb->msi); 2141 if (!sphb->msi_devs_num) { 2142 return 0; 2143 } 2144 sphb->msi_devs = g_new(spapr_pci_msi_mig, sphb->msi_devs_num); 2145 2146 g_hash_table_iter_init(&iter, sphb->msi); 2147 for (i = 0; g_hash_table_iter_next(&iter, &key, &value); ++i) { 2148 sphb->msi_devs[i].key = *(uint32_t *) key; 2149 sphb->msi_devs[i].value = *(spapr_pci_msi *) value; 2150 } 2151 2152 return 0; 2153 } 2154 2155 static int spapr_pci_post_load(void *opaque, int version_id) 2156 { 2157 SpaprPhbState *sphb = opaque; 2158 gpointer key, value; 2159 int i; 2160 2161 for (i = 0; i < sphb->msi_devs_num; ++i) { 2162 key = g_memdup(&sphb->msi_devs[i].key, 2163 sizeof(sphb->msi_devs[i].key)); 2164 value = g_memdup(&sphb->msi_devs[i].value, 2165 sizeof(sphb->msi_devs[i].value)); 2166 g_hash_table_insert(sphb->msi, key, value); 2167 } 2168 g_free(sphb->msi_devs); 2169 sphb->msi_devs = NULL; 2170 sphb->msi_devs_num = 0; 2171 2172 return 0; 2173 } 2174 2175 static bool pre_2_8_migration(void *opaque, int version_id) 2176 { 2177 SpaprPhbState *sphb = opaque; 2178 2179 return sphb->pre_2_8_migration; 2180 } 2181 2182 static const VMStateDescription vmstate_spapr_pci = { 2183 .name = "spapr_pci", 2184 .version_id = 2, 2185 .minimum_version_id = 2, 2186 .pre_save = spapr_pci_pre_save, 2187 .post_load = spapr_pci_post_load, 2188 .fields = (VMStateField[]) { 2189 VMSTATE_UINT64_EQUAL(buid, SpaprPhbState, NULL), 2190 VMSTATE_UINT32_TEST(mig_liobn, SpaprPhbState, pre_2_8_migration), 2191 VMSTATE_UINT64_TEST(mig_mem_win_addr, SpaprPhbState, pre_2_8_migration), 2192 VMSTATE_UINT64_TEST(mig_mem_win_size, SpaprPhbState, pre_2_8_migration), 2193 VMSTATE_UINT64_TEST(mig_io_win_addr, SpaprPhbState, pre_2_8_migration), 2194 VMSTATE_UINT64_TEST(mig_io_win_size, SpaprPhbState, pre_2_8_migration), 2195 VMSTATE_STRUCT_ARRAY(lsi_table, SpaprPhbState, PCI_NUM_PINS, 0, 2196 vmstate_spapr_pci_lsi, struct spapr_pci_lsi), 2197 VMSTATE_INT32(msi_devs_num, SpaprPhbState), 2198 VMSTATE_STRUCT_VARRAY_ALLOC(msi_devs, SpaprPhbState, msi_devs_num, 0, 2199 vmstate_spapr_pci_msi, spapr_pci_msi_mig), 2200 VMSTATE_END_OF_LIST() 2201 }, 2202 }; 2203 2204 static const char *spapr_phb_root_bus_path(PCIHostState *host_bridge, 2205 PCIBus *rootbus) 2206 { 2207 SpaprPhbState *sphb = SPAPR_PCI_HOST_BRIDGE(host_bridge); 2208 2209 return sphb->dtbusname; 2210 } 2211 2212 static void spapr_phb_class_init(ObjectClass *klass, void *data) 2213 { 2214 PCIHostBridgeClass *hc = PCI_HOST_BRIDGE_CLASS(klass); 2215 DeviceClass *dc = DEVICE_CLASS(klass); 2216 HotplugHandlerClass *hp = HOTPLUG_HANDLER_CLASS(klass); 2217 2218 hc->root_bus_path = spapr_phb_root_bus_path; 2219 dc->realize = spapr_phb_realize; 2220 dc->unrealize = spapr_phb_unrealize; 2221 dc->props = spapr_phb_properties; 2222 dc->reset = spapr_phb_reset; 2223 dc->vmsd = &vmstate_spapr_pci; 2224 /* Supported by TYPE_SPAPR_MACHINE */ 2225 dc->user_creatable = true; 2226 set_bit(DEVICE_CATEGORY_BRIDGE, dc->categories); 2227 hp->plug = spapr_pci_plug; 2228 hp->unplug = spapr_pci_unplug; 2229 hp->unplug_request = spapr_pci_unplug_request; 2230 } 2231 2232 static const TypeInfo spapr_phb_info = { 2233 .name = TYPE_SPAPR_PCI_HOST_BRIDGE, 2234 .parent = TYPE_PCI_HOST_BRIDGE, 2235 .instance_size = sizeof(SpaprPhbState), 2236 .instance_finalize = spapr_phb_finalizefn, 2237 .class_init = spapr_phb_class_init, 2238 .interfaces = (InterfaceInfo[]) { 2239 { TYPE_HOTPLUG_HANDLER }, 2240 { } 2241 } 2242 }; 2243 2244 static void spapr_phb_pci_enumerate_bridge(PCIBus *bus, PCIDevice *pdev, 2245 void *opaque) 2246 { 2247 unsigned int *bus_no = opaque; 2248 PCIBus *sec_bus = NULL; 2249 2250 if ((pci_default_read_config(pdev, PCI_HEADER_TYPE, 1) != 2251 PCI_HEADER_TYPE_BRIDGE)) { 2252 return; 2253 } 2254 2255 (*bus_no)++; 2256 pci_default_write_config(pdev, PCI_PRIMARY_BUS, pci_dev_bus_num(pdev), 1); 2257 pci_default_write_config(pdev, PCI_SECONDARY_BUS, *bus_no, 1); 2258 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1); 2259 2260 sec_bus = pci_bridge_get_sec_bus(PCI_BRIDGE(pdev)); 2261 if (!sec_bus) { 2262 return; 2263 } 2264 2265 pci_for_each_device(sec_bus, pci_bus_num(sec_bus), 2266 spapr_phb_pci_enumerate_bridge, bus_no); 2267 pci_default_write_config(pdev, PCI_SUBORDINATE_BUS, *bus_no, 1); 2268 } 2269 2270 static void spapr_phb_pci_enumerate(SpaprPhbState *phb) 2271 { 2272 PCIBus *bus = PCI_HOST_BRIDGE(phb)->bus; 2273 unsigned int bus_no = 0; 2274 2275 pci_for_each_device(bus, pci_bus_num(bus), 2276 spapr_phb_pci_enumerate_bridge, 2277 &bus_no); 2278 2279 } 2280 2281 int spapr_dt_phb(SpaprPhbState *phb, uint32_t intc_phandle, void *fdt, 2282 uint32_t nr_msis, int *node_offset) 2283 { 2284 int bus_off, i, j, ret; 2285 uint32_t bus_range[] = { cpu_to_be32(0), cpu_to_be32(0xff) }; 2286 struct { 2287 uint32_t hi; 2288 uint64_t child; 2289 uint64_t parent; 2290 uint64_t size; 2291 } QEMU_PACKED ranges[] = { 2292 { 2293 cpu_to_be32(b_ss(1)), cpu_to_be64(0), 2294 cpu_to_be64(phb->io_win_addr), 2295 cpu_to_be64(memory_region_size(&phb->iospace)), 2296 }, 2297 { 2298 cpu_to_be32(b_ss(2)), cpu_to_be64(SPAPR_PCI_MEM_WIN_BUS_OFFSET), 2299 cpu_to_be64(phb->mem_win_addr), 2300 cpu_to_be64(phb->mem_win_size), 2301 }, 2302 { 2303 cpu_to_be32(b_ss(3)), cpu_to_be64(phb->mem64_win_pciaddr), 2304 cpu_to_be64(phb->mem64_win_addr), 2305 cpu_to_be64(phb->mem64_win_size), 2306 }, 2307 }; 2308 const unsigned sizeof_ranges = 2309 (phb->mem64_win_size ? 3 : 2) * sizeof(ranges[0]); 2310 uint64_t bus_reg[] = { cpu_to_be64(phb->buid), 0 }; 2311 uint32_t interrupt_map_mask[] = { 2312 cpu_to_be32(b_ddddd(-1)|b_fff(0)), 0x0, 0x0, cpu_to_be32(-1)}; 2313 uint32_t interrupt_map[PCI_SLOT_MAX * PCI_NUM_PINS][7]; 2314 uint32_t ddw_applicable[] = { 2315 cpu_to_be32(RTAS_IBM_QUERY_PE_DMA_WINDOW), 2316 cpu_to_be32(RTAS_IBM_CREATE_PE_DMA_WINDOW), 2317 cpu_to_be32(RTAS_IBM_REMOVE_PE_DMA_WINDOW) 2318 }; 2319 uint32_t ddw_extensions[] = { 2320 cpu_to_be32(1), 2321 cpu_to_be32(RTAS_IBM_RESET_PE_DMA_WINDOW) 2322 }; 2323 uint32_t associativity[] = {cpu_to_be32(0x4), 2324 cpu_to_be32(0x0), 2325 cpu_to_be32(0x0), 2326 cpu_to_be32(0x0), 2327 cpu_to_be32(phb->numa_node)}; 2328 SpaprTceTable *tcet; 2329 SpaprDrc *drc; 2330 Error *errp = NULL; 2331 2332 /* Start populating the FDT */ 2333 _FDT(bus_off = fdt_add_subnode(fdt, 0, phb->dtbusname)); 2334 if (node_offset) { 2335 *node_offset = bus_off; 2336 } 2337 2338 /* Write PHB properties */ 2339 _FDT(fdt_setprop_string(fdt, bus_off, "device_type", "pci")); 2340 _FDT(fdt_setprop_string(fdt, bus_off, "compatible", "IBM,Logical_PHB")); 2341 _FDT(fdt_setprop_cell(fdt, bus_off, "#interrupt-cells", 0x1)); 2342 _FDT(fdt_setprop(fdt, bus_off, "used-by-rtas", NULL, 0)); 2343 _FDT(fdt_setprop(fdt, bus_off, "bus-range", &bus_range, sizeof(bus_range))); 2344 _FDT(fdt_setprop(fdt, bus_off, "ranges", &ranges, sizeof_ranges)); 2345 _FDT(fdt_setprop(fdt, bus_off, "reg", &bus_reg, sizeof(bus_reg))); 2346 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pci-config-space-type", 0x1)); 2347 _FDT(fdt_setprop_cell(fdt, bus_off, "ibm,pe-total-#msi", nr_msis)); 2348 2349 /* Dynamic DMA window */ 2350 if (phb->ddw_enabled) { 2351 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-applicable", &ddw_applicable, 2352 sizeof(ddw_applicable))); 2353 _FDT(fdt_setprop(fdt, bus_off, "ibm,ddw-extensions", 2354 &ddw_extensions, sizeof(ddw_extensions))); 2355 } 2356 2357 /* Advertise NUMA via ibm,associativity */ 2358 if (phb->numa_node != -1) { 2359 _FDT(fdt_setprop(fdt, bus_off, "ibm,associativity", associativity, 2360 sizeof(associativity))); 2361 } 2362 2363 /* Build the interrupt-map, this must matches what is done 2364 * in pci_swizzle_map_irq_fn 2365 */ 2366 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map-mask", 2367 &interrupt_map_mask, sizeof(interrupt_map_mask))); 2368 for (i = 0; i < PCI_SLOT_MAX; i++) { 2369 for (j = 0; j < PCI_NUM_PINS; j++) { 2370 uint32_t *irqmap = interrupt_map[i*PCI_NUM_PINS + j]; 2371 int lsi_num = pci_swizzle(i, j); 2372 2373 irqmap[0] = cpu_to_be32(b_ddddd(i)|b_fff(0)); 2374 irqmap[1] = 0; 2375 irqmap[2] = 0; 2376 irqmap[3] = cpu_to_be32(j+1); 2377 irqmap[4] = cpu_to_be32(intc_phandle); 2378 spapr_dt_irq(&irqmap[5], phb->lsi_table[lsi_num].irq, true); 2379 } 2380 } 2381 /* Write interrupt map */ 2382 _FDT(fdt_setprop(fdt, bus_off, "interrupt-map", &interrupt_map, 2383 sizeof(interrupt_map))); 2384 2385 tcet = spapr_tce_find_by_liobn(phb->dma_liobn[0]); 2386 if (!tcet) { 2387 return -1; 2388 } 2389 spapr_dma_dt(fdt, bus_off, "ibm,dma-window", 2390 tcet->liobn, tcet->bus_offset, 2391 tcet->nb_table << tcet->page_shift); 2392 2393 drc = spapr_drc_by_id(TYPE_SPAPR_DRC_PHB, phb->index); 2394 if (drc) { 2395 uint32_t drc_index = cpu_to_be32(spapr_drc_index(drc)); 2396 2397 _FDT(fdt_setprop(fdt, bus_off, "ibm,my-drc-index", &drc_index, 2398 sizeof(drc_index))); 2399 } 2400 2401 /* Walk the bridges and program the bus numbers*/ 2402 spapr_phb_pci_enumerate(phb); 2403 _FDT(fdt_setprop_cell(fdt, bus_off, "qemu,phb-enumerated", 0x1)); 2404 2405 /* Walk the bridge and subordinate buses */ 2406 ret = spapr_dt_pci_bus(phb, PCI_HOST_BRIDGE(phb)->bus, fdt, bus_off); 2407 if (ret < 0) { 2408 return ret; 2409 } 2410 2411 spapr_phb_nvgpu_populate_dt(phb, fdt, bus_off, &errp); 2412 if (errp) { 2413 error_report_err(errp); 2414 } 2415 spapr_phb_nvgpu_ram_populate_dt(phb, fdt); 2416 2417 return 0; 2418 } 2419 2420 void spapr_pci_rtas_init(void) 2421 { 2422 spapr_rtas_register(RTAS_READ_PCI_CONFIG, "read-pci-config", 2423 rtas_read_pci_config); 2424 spapr_rtas_register(RTAS_WRITE_PCI_CONFIG, "write-pci-config", 2425 rtas_write_pci_config); 2426 spapr_rtas_register(RTAS_IBM_READ_PCI_CONFIG, "ibm,read-pci-config", 2427 rtas_ibm_read_pci_config); 2428 spapr_rtas_register(RTAS_IBM_WRITE_PCI_CONFIG, "ibm,write-pci-config", 2429 rtas_ibm_write_pci_config); 2430 if (msi_nonbroken) { 2431 spapr_rtas_register(RTAS_IBM_QUERY_INTERRUPT_SOURCE_NUMBER, 2432 "ibm,query-interrupt-source-number", 2433 rtas_ibm_query_interrupt_source_number); 2434 spapr_rtas_register(RTAS_IBM_CHANGE_MSI, "ibm,change-msi", 2435 rtas_ibm_change_msi); 2436 } 2437 2438 spapr_rtas_register(RTAS_IBM_SET_EEH_OPTION, 2439 "ibm,set-eeh-option", 2440 rtas_ibm_set_eeh_option); 2441 spapr_rtas_register(RTAS_IBM_GET_CONFIG_ADDR_INFO2, 2442 "ibm,get-config-addr-info2", 2443 rtas_ibm_get_config_addr_info2); 2444 spapr_rtas_register(RTAS_IBM_READ_SLOT_RESET_STATE2, 2445 "ibm,read-slot-reset-state2", 2446 rtas_ibm_read_slot_reset_state2); 2447 spapr_rtas_register(RTAS_IBM_SET_SLOT_RESET, 2448 "ibm,set-slot-reset", 2449 rtas_ibm_set_slot_reset); 2450 spapr_rtas_register(RTAS_IBM_CONFIGURE_PE, 2451 "ibm,configure-pe", 2452 rtas_ibm_configure_pe); 2453 spapr_rtas_register(RTAS_IBM_SLOT_ERROR_DETAIL, 2454 "ibm,slot-error-detail", 2455 rtas_ibm_slot_error_detail); 2456 } 2457 2458 static void spapr_pci_register_types(void) 2459 { 2460 type_register_static(&spapr_phb_info); 2461 } 2462 2463 type_init(spapr_pci_register_types) 2464 2465 static int spapr_switch_one_vga(DeviceState *dev, void *opaque) 2466 { 2467 bool be = *(bool *)opaque; 2468 2469 if (object_dynamic_cast(OBJECT(dev), "VGA") 2470 || object_dynamic_cast(OBJECT(dev), "secondary-vga")) { 2471 object_property_set_bool(OBJECT(dev), be, "big-endian-framebuffer", 2472 &error_abort); 2473 } 2474 return 0; 2475 } 2476 2477 void spapr_pci_switch_vga(bool big_endian) 2478 { 2479 SpaprMachineState *spapr = SPAPR_MACHINE(qdev_get_machine()); 2480 SpaprPhbState *sphb; 2481 2482 /* 2483 * For backward compatibility with existing guests, we switch 2484 * the endianness of the VGA controller when changing the guest 2485 * interrupt mode 2486 */ 2487 QLIST_FOREACH(sphb, &spapr->phbs, list) { 2488 BusState *bus = &PCI_HOST_BRIDGE(sphb)->bus->qbus; 2489 qbus_walk_children(bus, spapr_switch_one_vga, NULL, NULL, NULL, 2490 &big_endian); 2491 } 2492 } 2493